We try to understand two following issues in more systematic ways.

What happens if keywords (seed) are mixed across topics?
What happens if good keywords and bad keywords are mixed?

Some terminology used in this report.

Strong seed: chosen by the most frequently words within each topic (0.9% to 1.5%)
Medium quality seed: chosen by the words which appears about 0.1% in the whole corpus
Weak seed: chosen by the words which appears about 0.05% in the whole corpus

For the number of topics, we focus on \(K + 1 (=4)\).
Description of data

Parameters	Value
# Document	\(1000\)
Length of documents	Poisson( \(200\) )
\(\alpha\) (control topic)	(0.953, 0.644, 1.091, 0.508, 1.216, 0.741)

When we look at the vector \(\alpha\), we learn that words in topic 2 appear less frequent then other topics. This could be the reason that estimation of topic 2 is sometimes instable in the following results.

Prepare the corpus

docs <- list.files(folder, pattern = "*.txt", full.names = TRUE)
explore_ <- explore(docs,
             remove_numbers = FALSE, # For simulation, make it false
             remove_punct = TRUE,
             remove_symbols = TRUE,
             remove_separators = TRUE)
explore_$data_tfidf %>%
    group_by(term) %>%
    summarize(TFIDFMedian = median(tf_idf)) %>%
    arrange(desc(TFIDFMedian)) -> ranked_tfidf

Mix seeds (keywords)

We include one or two seeds from other topic (topic 3) in the keyword set of topic 1.

Result 1: two strong seeds from topic 1 and one strong seed from topic 3

seed_list <- list(c("w1t1 w367t1 w80t3"),
                c("w223t2 w541t2 w536t2"),
                c("w26t3 w6t3 w223t3"))
explore_$visualize_dict_prop(seed_list)

top_terms(contam1, n = 25, show_seed = T)

##       1            2            3            T_1     
##  [1,] "w1t1 [✓]"   "w223t2 [✓]" "w26t3 [✓]"  "w33t5" 
##  [2,] "w367t1 [✓]" "w541t2 [✓]" "w6t3 [✓]"   "w65t5" 
##  [3,] "w180t1"     "w536t2 [✓]" "w223t3 [✓]" "w173t6"
##  [4,] "w80t3 [✓]"  "w0t2"       "w8t3"       "w118t5"
##  [5,] "w72t1"      "w1245t2"    "w87t3"      "w19t5" 
##  [6,] "w75t1"      "w122t2"     "w77t3"      "w45t6" 
##  [7,] "w63t1"      "w164t2"     "w127t3"     "w651t6"
##  [8,] "w449t1"     "w619t2"     "w264t3"     "w185t6"
##  [9,] "w60t1"      "w23t2"      "w153t3"     "w55t6" 
## [10,] "w832t1"     "w725t2"     "w115t3"     "w58t5" 
## [11,] "w11t1"      "w623t2"     "w143t3"     "w37t6" 
## [12,] "w405t1"     "w943t2"     "w263t3"     "w209t5"
## [13,] "w1073t1"    "w979t2"     "w92t3"      "w107t5"
## [14,] "w833t1"     "w606t2"     "w867t3"     "w66t5" 
## [15,] "w113t1"     "w1501t2"    "w593t3"     "w129t5"
## [16,] "w273t1"     "w158t2"     "w949t3"     "w33t4" 
## [17,] "w404t1"     "w134t2"     "w222t3"     "w368t5"
## [18,] "w21t1"      "w724t2"     "w884t3"     "w269t5"
## [19,] "w403t1"     "w556t2"     "w953t3"     "w91t5" 
## [20,] "w808t1"     "w568t2"     "w196t3"     "w116t5"
## [21,] "w1214t1"    "w129t2"     "w548t3"     "w379t5"
## [22,] "w274t1"     "w886t2"     "w848t3"     "w160t6"
## [23,] "w391t1"     "w159t2"     "w80t3 [1]"  "w400t5"
## [24,] "w768t1"     "w629t2"     "w245t3"     "w686t5"
## [25,] "w994t1"     "w170t2"     "w103t3"     "w430t5"

diagnosis_topic_recovery_heatmap(contam1, 25)
lda4

diagnosis_model_fit(contam1, start=2)

Result 2: two strong seeds from topic 1 and one middle quality seed from topic 3

seed_list <- list(c("w1t1 w367t1 w213t3"),
                c("w223t2 w541t2 w536t2"),
                c("w26t3 w6t3 w223t3"))
explore_$visualize_dict_prop(seed_list)

top_terms(contam2, show_seed = T)

##       1            2            3            T_1     
##  [1,] "w1t1 [✓]"   "w223t2 [✓]" "w26t3 [✓]"  "w33t5" 
##  [2,] "w367t1 [✓]" "w33t4"      "w6t3 [✓]"   "w65t5" 
##  [3,] "w180t1"     "w541t2 [✓]" "w223t3 [✓]" "w173t6"
##  [4,] "w72t1"      "w536t2 [✓]" "w80t3"      "w118t5"
##  [5,] "w75t1"      "w0t2"       "w8t3"       "w19t5" 
##  [6,] "w63t1"      "w1245t2"    "w87t3"      "w45t6" 
##  [7,] "w832t1"     "w83t4"      "w77t3"      "w651t6"
##  [8,] "w449t1"     "w122t2"     "w127t3"     "w185t6"
##  [9,] "w60t1"      "w164t2"     "w264t3"     "w55t6" 
## [10,] "w11t1"      "w619t2"     "w153t3"     "w58t5"

diagnosis_topic_recovery_heatmap(contam2, 25)
lda4

diagnosis_model_fit(contam2, start=2)

Result 3: two strong seeds from topic 1 and one weak seed from other topic 3

seed_list <- list(c("w1t1 w367t1 w206t3"),
        c("w223t2 w541t2 w536t2"),
        c("w26t3 w6t3 w223t3"))
explore_$visualize_dict_prop(seed_list)

top_terms(contam3, show_seed = T)

##       1            2            3            T_1     
##  [1,] "w1t1 [✓]"   "w223t2 [✓]" "w26t3 [✓]"  "w33t5" 
##  [2,] "w367t1 [✓]" "w541t2 [✓]" "w6t3 [✓]"   "w65t5" 
##  [3,] "w180t1"     "w536t2 [✓]" "w223t3 [✓]" "w173t6"
##  [4,] "w33t4"      "w0t2"       "w80t3"      "w118t5"
##  [5,] "w72t1"      "w1245t2"    "w8t3"       "w19t5" 
##  [6,] "w75t1"      "w122t2"     "w87t3"      "w45t6" 
##  [7,] "w63t1"      "w164t2"     "w77t3"      "w651t6"
##  [8,] "w449t1"     "w619t2"     "w127t3"     "w185t6"
##  [9,] "w60t1"      "w23t2"      "w264t3"     "w55t6" 
## [10,] "w11t1"      "w725t2"     "w153t3"     "w58t5"

diagnosis_topic_recovery_heatmap(contam3, 25)
lda4

diagnosis_model_fit(contam3, start=2)

Result 4: one strong seed from topic 1 and two weak seed from other topic 3

seed_list <- list(c("w1t1 w213t3 w387t3"),
        c("w223t2 w541t2 w536t2"),
        c("w26t3 w6t3 w223t3"))
explore_$visualize_dict_prop(seed_list)

top_terms(contam4, show_seed = T)

##       1          2            3            T_1     
##  [1,] "w1t1 [✓]" "w223t2 [✓]" "w26t3 [✓]"  "w33t5" 
##  [2,] "w367t1"   "w541t2 [✓]" "w6t3 [✓]"   "w65t5" 
##  [3,] "w180t1"   "w536t2 [✓]" "w223t3 [✓]" "w173t6"
##  [4,] "w72t1"    "w0t2"       "w80t3"      "w118t5"
##  [5,] "w75t1"    "w1245t2"    "w8t3"       "w19t5" 
##  [6,] "w63t1"    "w122t2"     "w87t3"      "w45t6" 
##  [7,] "w832t1"   "w164t2"     "w77t3"      "w651t6"
##  [8,] "w449t1"   "w619t2"     "w127t3"     "w185t6"
##  [9,] "w60t1"    "w23t2"      "w264t3"     "w55t6" 
## [10,] "w11t1"    "w725t2"     "w153t3"     "w58t5"

diagnosis_topic_recovery_heatmap(contam4, 25)
lda4

diagnosis_model_fit(contam4, start=2)

Result 5: one strong seed from topic 1 and two weak seeds from other topic 3

seed_list <- list(c("w1t1 w1492t3 w206t3"),
        c("w223t2 w541t2 w536t2"),
        c("w26t3 w6t3 w223t3"))
explore_$visualize_dict_prop(seed_list)

top_terms(contam5, show_seed = T)

##       1          2            3            T_1     
##  [1,] "w1t1 [✓]" "w223t2 [✓]" "w26t3 [✓]"  "w33t5" 
##  [2,] "w367t1"   "w33t4"      "w6t3 [✓]"   "w65t5" 
##  [3,] "w180t1"   "w541t2 [✓]" "w223t3 [✓]" "w173t6"
##  [4,] "w72t1"    "w536t2 [✓]" "w80t3"      "w118t5"
##  [5,] "w75t1"    "w0t2"       "w8t3"       "w19t5" 
##  [6,] "w63t1"    "w1245t2"    "w87t3"      "w45t6" 
##  [7,] "w449t1"   "w83t4"      "w77t3"      "w651t6"
##  [8,] "w832t1"   "w122t2"     "w127t3"     "w55t6" 
##  [9,] "w60t1"    "w164t2"     "w264t3"     "w185t6"
## [10,] "w11t1"    "w619t2"     "w153t3"     "w58t5"

diagnosis_topic_recovery_heatmap(contam5, 25)
lda4

diagnosis_model_fit(contam5, start=2)

Result 6: two middle quality seeds from topic 1 and one middle quality seed from other topic 3

seed_list <- list(c("w1281t1 w2461t1 w213t3"),
        c("w223t2 w541t2 w536t2"),
        c("w26t3 w6t3 w223t3"))
explore_$visualize_dict_prop(seed_list)

top_terms(contam6, show_seed = T)

##       1        2            3            T_1     
##  [1,] "w33t5"  "w223t2 [✓]" "w26t3 [✓]"  "w173t6"
##  [2,] "w65t5"  "w541t2 [✓]" "w6t3 [✓]"   "w45t6" 
##  [3,] "w118t5" "w536t2 [✓]" "w223t3 [✓]" "w651t6"
##  [4,] "w1t1"   "w0t2"       "w80t3"      "w185t6"
##  [5,] "w19t5"  "w1245t2"    "w8t3"       "w55t6" 
##  [6,] "w367t1" "w122t2"     "w87t3"      "w37t6" 
##  [7,] "w180t1" "w164t2"     "w77t3"      "w33t4" 
##  [8,] "w58t5"  "w619t2"     "w127t3"     "w160t6"
##  [9,] "w209t5" "w23t2"      "w264t3"     "w83t4" 
## [10,] "w107t5" "w725t2"     "w153t3"     "w50t6"

diagnosis_topic_recovery_heatmap(contam6, 25, topicvec = c(1,2,3,4))
lda4

diagnosis_model_fit(contam6, start=2)

Result 7: two middle quality seeds from topic 1 and one weak seed from other topic 3

seed_list <- list(c("w1281t1 w2461t1 w206t3"),
        c("w223t2 w541t2 w536t2"),
        c("w26t3 w6t3 w223t3"))
explore_$visualize_dict_prop(seed_list)

top_terms(contam7, show_seed = T)

##       1        2            3            T_1     
##  [1,] "w33t5"  "w223t2 [✓]" "w26t3 [✓]"  "w173t6"
##  [2,] "w65t5"  "w541t2 [✓]" "w6t3 [✓]"   "w45t6" 
##  [3,] "w118t5" "w536t2 [✓]" "w223t3 [✓]" "w651t6"
##  [4,] "w1t1"   "w0t2"       "w80t3"      "w185t6"
##  [5,] "w19t5"  "w1245t2"    "w8t3"       "w55t6" 
##  [6,] "w367t1" "w122t2"     "w87t3"      "w37t6" 
##  [7,] "w180t1" "w164t2"     "w77t3"      "w33t4" 
##  [8,] "w209t5" "w619t2"     "w127t3"     "w160t6"
##  [9,] "w107t5" "w23t2"      "w264t3"     "w83t4" 
## [10,] "w58t5"  "w725t2"     "w153t3"     "w50t6"

diagnosis_topic_recovery_heatmap(contam7, 25, topicvec = c(1,2,3,4))
lda4

diagnosis_model_fit(contam7, start=2)

Mix of strong, medium quality, and weak seeds within one topic

We only change topic 1, holding seeds of topic 2 and topic 3 constant (all strong seeds). ## Result 0: baseline, all strong seeds

seed_list <- list(c("w1t1 w367t1 w180t1"),
                c("w223t2 w541t2 w536t2"),
                c("w26t3 w6t3 w223t3"))
explore_$visualize_dict_prop(seed_list)

explore_$visualize_tfidf(seed_list)

## $density

## 
## $median

top_terms(tf1, show_seed = T)

##       1            2            3            T_1     
##  [1,] "w1t1 [✓]"   "w173t6"     "w26t3 [✓]"  "w33t5" 
##  [2,] "w367t1 [✓]" "w45t6"      "w6t3 [✓]"   "w65t5" 
##  [3,] "w180t1 [✓]" "w651t6"     "w223t3 [✓]" "w118t5"
##  [4,] "w72t1"      "w185t6"     "w80t3"      "w19t5" 
##  [5,] "w75t1"      "w223t2 [✓]" "w8t3"       "w58t5" 
##  [6,] "w63t1"      "w55t6"      "w87t3"      "w209t5"
##  [7,] "w449t1"     "w37t6"      "w77t3"      "w107t5"
##  [8,] "w832t1"     "w160t6"     "w127t3"     "w66t5" 
##  [9,] "w60t1"      "w541t2 [✓]" "w264t3"     "w129t5"
## [10,] "w11t1"      "w536t2 [✓]" "w153t3"     "w368t5"

diagnosis_topic_recovery_heatmap(tf1, 25)
lda4

diagnosis_model_fit(tf1, start=2)

Result 1: two strong seeds and one middle quality seed

seed_list <- list(c("w1t1 w367t1 w1671t1"),
                c("w223t2 w541t2 w536t2"),
                c("w26t3 w6t3 w223t3"))
explore_$visualize_dict_prop(seed_list)

top_terms(mix1, show_seed = T)

##       1            2            3            T_1     
##  [1,] "w1t1 [✓]"   "w173t6"     "w26t3 [✓]"  "w33t5" 
##  [2,] "w367t1 [✓]" "w45t6"      "w6t3 [✓]"   "w65t5" 
##  [3,] "w180t1"     "w651t6"     "w223t3 [✓]" "w118t5"
##  [4,] "w72t1"      "w185t6"     "w80t3"      "w19t5" 
##  [5,] "w75t1"      "w223t2 [✓]" "w8t3"       "w58t5" 
##  [6,] "w63t1"      "w55t6"      "w87t3"      "w209t5"
##  [7,] "w832t1"     "w37t6"      "w77t3"      "w107t5"
##  [8,] "w449t1"     "w160t6"     "w127t3"     "w66t5" 
##  [9,] "w60t1"      "w541t2 [✓]" "w264t3"     "w129t5"
## [10,] "w11t1"      "w536t2 [✓]" "w153t3"     "w33t4"

diagnosis_topic_recovery_heatmap(mix1, 25)
lda4

diagnosis_model_fit(mix1, start=2)

Result 2: two strong seeds one weak seed

seed_list <- list(c("w1t1 w367t1 w2532t1"),
                c("w223t2 w541t2 w536t2"),
                c("w26t3 w6t3 w223t3"))
explore_$visualize_dict_prop(seed_list)

top_terms(mix2, show_seed = T)

##       1            2            3            T_1     
##  [1,] "w1t1 [✓]"   "w223t2 [✓]" "w26t3 [✓]"  "w33t5" 
##  [2,] "w367t1 [✓]" "w541t2 [✓]" "w6t3 [✓]"   "w65t5" 
##  [3,] "w180t1"     "w536t2 [✓]" "w223t3 [✓]" "w173t6"
##  [4,] "w72t1"      "w0t2"       "w80t3"      "w118t5"
##  [5,] "w75t1"      "w1245t2"    "w8t3"       "w19t5" 
##  [6,] "w63t1"      "w122t2"     "w87t3"      "w45t6" 
##  [7,] "w832t1"     "w164t2"     "w77t3"      "w185t6"
##  [8,] "w449t1"     "w619t2"     "w127t3"     "w651t6"
##  [9,] "w60t1"      "w23t2"      "w264t3"     "w55t6" 
## [10,] "w11t1"      "w725t2"     "w153t3"     "w58t5"

diagnosis_topic_recovery_heatmap(mix2, 25)
lda4

diagnosis_model_fit(mix2, start=2)

Result 3: one strong seed and two middle quality seeds

seed_list <- list(c("w1t1 w1185t1 w1671t1"),
        c("w223t2 w541t2 w536t2"),
        c("w26t3 w6t3 w223t3"))
explore_$visualize_dict_prop(seed_list)

top_terms(mix3, show_seed = T)

##       1          2            3            T_1     
##  [1,] "w1t1 [✓]" "w223t2 [✓]" "w26t3 [✓]"  "w33t5" 
##  [2,] "w367t1"   "w541t2 [✓]" "w6t3 [✓]"   "w65t5" 
##  [3,] "w180t1"   "w536t2 [✓]" "w223t3 [✓]" "w173t6"
##  [4,] "w72t1"    "w0t2"       "w80t3"      "w118t5"
##  [5,] "w75t1"    "w1245t2"    "w8t3"       "w19t5" 
##  [6,] "w63t1"    "w122t2"     "w87t3"      "w45t6" 
##  [7,] "w449t1"   "w164t2"     "w77t3"      "w651t6"
##  [8,] "w832t1"   "w619t2"     "w127t3"     "w185t6"
##  [9,] "w11t1"    "w23t2"      "w264t3"     "w55t6" 
## [10,] "w405t1"   "w725t2"     "w153t3"     "w58t5"

diagnosis_topic_recovery_heatmap(mix3, 25)
lda4

diagnosis_model_fit(mix3, start=2)

Result 4: one strong seed and two weak seeds

seed_list <- list(c("w1t1 w2461t1 w2532t1"),
        c("w223t2 w541t2 w536t2"),
        c("w26t3 w6t3 w223t3"))
explore_$visualize_dict_prop(seed_list)

top_terms(mix4, show_seed = T)

##       1          2            3            T_1     
##  [1,] "w1t1 [✓]" "w173t6"     "w26t3 [✓]"  "w33t5" 
##  [2,] "w367t1"   "w45t6"      "w6t3 [✓]"   "w65t5" 
##  [3,] "w180t1"   "w651t6"     "w223t3 [✓]" "w118t5"
##  [4,] "w72t1"    "w185t6"     "w80t3"      "w19t5" 
##  [5,] "w75t1"    "w223t2 [✓]" "w8t3"       "w58t5" 
##  [6,] "w63t1"    "w55t6"      "w87t3"      "w209t5"
##  [7,] "w832t1"   "w37t6"      "w77t3"      "w107t5"
##  [8,] "w449t1"   "w160t6"     "w127t3"     "w66t5" 
##  [9,] "w60t1"    "w541t2 [✓]" "w264t3"     "w129t5"
## [10,] "w11t1"    "w536t2 [✓]" "w153t3"     "w33t4"

diagnosis_topic_recovery_heatmap(mix4, 25)
lda4

diagnosis_model_fit(mix4, start=2)

Result 5: two middle quality seeds a weak seed

seed_list <- list(c("w1281t1 w2461t1 w2532t1"),
        c("w223t2 w541t2 w536t2"),
        c("w26t3 w6t3 w223t3"))
explore_$visualize_dict_prop(seed_list)

top_terms(mix5, show_seed = T)

##       1        2            3            T_1     
##  [1,] "w1t1"   "w173t6"     "w26t3 [✓]"  "w33t5" 
##  [2,] "w367t1" "w45t6"      "w6t3 [✓]"   "w65t5" 
##  [3,] "w180t1" "w651t6"     "w223t3 [✓]" "w118t5"
##  [4,] "w33t4"  "w185t6"     "w80t3"      "w19t5" 
##  [5,] "w72t1"  "w223t2 [✓]" "w8t3"       "w58t5" 
##  [6,] "w75t1"  "w55t6"      "w87t3"      "w209t5"
##  [7,] "w63t1"  "w37t6"      "w77t3"      "w107t5"
##  [8,] "w832t1" "w541t2 [✓]" "w127t3"     "w66t5" 
##  [9,] "w449t1" "w160t6"     "w264t3"     "w129t5"
## [10,] "w60t1"  "w536t2 [✓]" "w153t3"     "w368t5"

diagnosis_topic_recovery_heatmap(mix5, 25)
lda4

diagnosis_model_fit(mix5, start=2)

Result 6: one middle quality seed and two week seed

seed_list <- list(c("w1281t1 w1185t1 w2532t1"),
        c("w223t2 w541t2 w536t2"),
        c("w26t3 w6t3 w223t3"))
explore_$visualize_dict_prop(seed_list)

top_terms(mix6, show_seed = T)

##       1        2            3            T_1     
##  [1,] "w173t6" "w223t2 [✓]" "w26t3 [✓]"  "w33t5" 
##  [2,] "w1t1"   "w541t2 [✓]" "w6t3 [✓]"   "w65t5" 
##  [3,] "w45t6"  "w536t2 [✓]" "w223t3 [✓]" "w118t5"
##  [4,] "w367t1" "w0t2"       "w80t3"      "w19t5" 
##  [5,] "w651t6" "w1245t2"    "w8t3"       "w58t5" 
##  [6,] "w185t6" "w122t2"     "w87t3"      "w209t5"
##  [7,] "w180t1" "w164t2"     "w77t3"      "w107t5"
##  [8,] "w37t6"  "w619t2"     "w127t3"     "w66t5" 
##  [9,] "w55t6"  "w23t2"      "w264t3"     "w129t5"
## [10,] "w72t1"  "w725t2"     "w153t3"     "w33t4"

diagnosis_topic_recovery_heatmap(mix6, 25, topicvec = c(1,2,3,4))
lda4

diagnosis_model_fit(mix6, start=2)

Result 7: one strong, one middle quality, one weak seeds

seed_list <- list(c("w1t1 w1185t1 w2532t1"),
        c("w223t2 w541t2 w536t2"),
        c("w26t3 w6t3 w223t3"))
explore_$visualize_dict_prop(seed_list)

top_terms(mix7, show_seed = T)

##       1          2            3            T_1     
##  [1,] "w1t1 [✓]" "w173t6"     "w26t3 [✓]"  "w33t5" 
##  [2,] "w367t1"   "w45t6"      "w6t3 [✓]"   "w65t5" 
##  [3,] "w180t1"   "w651t6"     "w223t3 [✓]" "w118t5"
##  [4,] "w72t1"    "w185t6"     "w80t3"      "w19t5" 
##  [5,] "w75t1"    "w223t2 [✓]" "w8t3"       "w58t5" 
##  [6,] "w63t1"    "w55t6"      "w87t3"      "w209t5"
##  [7,] "w832t1"   "w37t6"      "w77t3"      "w107t5"
##  [8,] "w449t1"   "w33t4"      "w127t3"     "w66t5" 
##  [9,] "w60t1"    "w541t2 [✓]" "w264t3"     "w129t5"
## [10,] "w11t1"    "w160t6"     "w153t3"     "w368t5"

diagnosis_topic_recovery_heatmap(mix7, 25, topicvec = c(1,2,3,4))
lda4

diagnosis_model_fit(mix7, start=2)

Result 8: three week seeds

seed_list <- list(c("w2132t1 w2461t1 w2532t1"),
          c("w223t2 w541t2 w536t2"),
          c("w26t3 w6t3 w223t3"))
explore_$visualize_dict_prop(seed_list)

explore_$visualize_tfidf(seed_list)

## $density

## 
## $median

top_terms(tf_sp, show_seed = T)

##       1        2            3            T_1     
##  [1,] "w33t5"  "w223t2 [✓]" "w26t3 [✓]"  "w173t6"
##  [2,] "w65t5"  "w541t2 [✓]" "w6t3 [✓]"   "w45t6" 
##  [3,] "w118t5" "w536t2 [✓]" "w223t3 [✓]" "w651t6"
##  [4,] "w1t1"   "w0t2"       "w80t3"      "w185t6"
##  [5,] "w19t5"  "w1245t2"    "w8t3"       "w55t6" 
##  [6,] "w367t1" "w122t2"     "w87t3"      "w37t6" 
##  [7,] "w180t1" "w164t2"     "w77t3"      "w33t4" 
##  [8,] "w58t5"  "w619t2"     "w127t3"     "w160t6"
##  [9,] "w209t5" "w23t2"      "w264t3"     "w83t4" 
## [10,] "w107t5" "w725t2"     "w153t3"     "w50t6"

diagnosis_topic_recovery_heatmap(tf_sp, 25, topicvec = c(1,2,3,4))
lda4

diagnosis_model_fit(tf_sp, start=2)

20180306report

Shusei Eshima/ Sasaki Tomoya

2018-03-06